INNER CODE UNIT · Python

tokenize_prompt

ASCIT31/Dark-Moon · mcp/src/server.py:141

def tokenize_prompt(text: str, session_id: Optional[str] = None) -> Dict[str, Any]:
    """
    Tokenize a launch prompt BEFORE it reaches the model.

    The Darkmoon privacy boundary normally acts only on tool calls and their
    output. The initial campaign prompt (TARGET / CREDS / SCOPE / TOKEN / IP /
    port) is sent to the model first, so without this it would reach the model in
    clear (issue #40, section 3). This applies the SAME per-session vault the
    tool calls use, so every placeholder is identical to the ones the gateway
    rehydrates during the run and restores into the local report — no value is
    ever hard-coded.

    Intended caller: the Darkmoon OpenCode privacy plugin, which rewrites the
    outgoing user message with the returned `tokenized` text before the model
    sees it. Safe to call repeatedly (deterministic per session).

    Args:
        text: the raw prompt text to tokenize.

View source record →